Skip to main content

Agent Fleet And Orchestration

Use the agent-fleet profile when a project coordinates multiple agents, workers, queues, workflows, SOAR playbooks, or distributed jobs.

This covers patterns such as:

  • LangGraph-style state graphs
  • Temporal-style durable workflows
  • Ray/Celery-style distributed task execution
  • queue and worker systems
  • MCP tool fleets
  • SOAR remediation playbooks
  • multi-agent pipelines

The goal is not to standardize on one orchestrator. The goal is to make the orchestration risk explicit before implementation.

Required design decisions:

  • workflow/state graph
  • durable state source of truth
  • agent and worker identity model
  • tool/action allowlist
  • retry, timeout, idempotency, and dead-letter behavior
  • backpressure and concurrency limits
  • observability and trace correlation
  • human approval boundaries
  • emergency pause or kill switch
  • rollback or compensation workflow

Recommended profile combinations:

sicario init my-agent-system --profile agent-fleet,supply-chain
sicario init my-soar-playbooks --profile agent-fleet,cloud-iac,compliance
sicario init my-ai-platform --profile appsec,ai-system,agent-fleet,supply-chain

For security automation, the important question is not only whether the agent can act. It is whether the action is bounded, approved, observable, reversible, and evidenced.